An Experimental Study of the Small World Problem Sociometry

advertisement
An Experimental Study of the Small World Problem
Jeffrey Travers; Stanley Milgram
Sociometry, Vol. 32, No. 4. (Dec., 1969), pp. 425-443.
Stable URL:
http://links.jstor.org/sici?sici=0038-0431%28196912%2932%3A4%3C425%3AAESOTS%3E2.0.CO%3B2-W
Sociometry is currently published by American Sociological Association.
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained
prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in
the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/journals/asa.html.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic
journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers,
and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take
advantage of advances in technology. For more information regarding JSTOR, please contact support@jstor.org.
http://www.jstor.org
Mon Sep 17 04:15:09 2007
An Experimental Study of the Small World Problem * JEFFREY TRAVERS
Harvard University
AND
STANLEY MILGRAM
The City University of New York
Arbitrarily selected individuals (N=296) i n Nebraska and Boston are asked
to genwate acquaintance chains t o a target person i n Massachusetts, employing "the small world method" (Milgram, 1967). Sixty-four chains reach
the target person. W i t h i n this group the mean number of intermediaries between starters and targets is 5.2. Boston starting chains reach the target
person with fewer intermediaries than those starting i n Nebraska; subpopulations i n the Nebraska group do not differ among themselves. T h e funneling
of chains through sociometric "stars" is noted, with 48 per cent of the chains
passing through three persons before reaching the target. Applications of the
method t o studies of large scale social structure are discussed.
The simplest way of formulating the small world problem is "what is the
probability that any two people, selected arbitrarily from a large population,
such as that of the United States, will know each other?" A more interesting
formulation, however, takes account of the fact that, while persons a and z
may not know each other directly, they may share one or more mutual
acquaintances; that is, there may exist a set of individuals, B , (consisting
of individuals 61, 62 . . . 6,) who know both a and z and thus link them
to one another: More generally, a and z may be connected not by any single
common acquaintance, but by a series of such intermediaries, a-b-c- . . -y-z;
i.e., a knows 6 (and no one else in the chain); 6 knows a and in addition
knows c, c in turn knows d, etc.
To elaborate the problem somewhat further, let us represent the popula-
.
* T h e study was carried out while both authors were a t Harvard University, and was
financed by grants from the Milton Fund and from the Harvard Laboratory of Social
Relations. Mr. Joseph Gerver provided invaluable assistance in summarizing and criticizing
the mathematical work discussed in this paper.
426
SOCIOMETRY
tion of the United States by a partially connected set of points. Let each
point represent a person, and let a line connecting two points signify that
the two individuals know each other. (Knowing is here assumed to be symmetric: if a knows b then b knows a. Substantively, "knowing" is used to
denote a mutual relationship; other senses of the verb, e.g. knowing about
a famous person, are excluded.) The structure takes the form of a cluster
of roughly 200 million points with a complex web of connections among
them. The acquaintance chains described above appear as pathways along
connected line segments. Unless some portion of the population is totally
isolated from the rest, such that no one in that subgroup knows anyone
outside it, there must be a t least one chain connecting any two people in the
population. In general there will be many such pathways, of various lengths,
between any two individuals.
In view of such a structure, one way of refining our statement of the
small world problem is the following: given two individuals selected randomly
from the population, what is the probability that the minimum number of
intermediaries required to link them is 0, 1, 2, . . . k ? (Alternatively, one
might ask not about the minimum chains between pairs of people, but
mean chain lengths, median chain lengths, etc.)
Perhaps the most direct way of attacking the small world problem is to
trace a number of real acquaintance chains in a large population. This is the
technique of the study reported in this paper. The phrase "small world"
suggests that social networks are in some sense tightly woven, full of unexpected strands linking individuals seemingly far removed from one another
in physical or social space. The principal question of the present investigation was whether such interconnectedness could be demonstrated experimentally.
The only example of mathematical treatment dealing directly with the
small world problem is the model provided by Ithiel Pool and Manfred
Kochen (unpublished manuscript). Pool and Kochen assume a population
of N individuals, each of whom knows, on the average, n others in the
population. They attempt to calculate Pk, the probability that two persons
chosen randomly from the group can be linked by a chain of k intermediaries.
Their basic model takes the form of a "tree" or geometric progression. Using
an estimate of average acquaintance volume provided by Gurevitch (1961))
they deduce that two intermediaries will be required to link typical pairs of
individuals in a population of 200 million. Their model does not take account
of social structure. Instead of allowing acquaintance nets to define the
boundaries of functioning social groups, Pool and Kochen must, for the purposes of their model, conceive of society as being partitioned into a number
of hypothetical groups, each with identical populations. They are then able
STUDY OF THE SMALL WORLD PROBLEM
to devise a way to predict chain lengths within and between such hypothesized
groups.
In an empirical study related to the small world problem Rapoport and
Horvath (1961) examined sociometric nets in a junior high school of 861
students. The authors asked students to name in order their eight best
friends within the school. They then traced the acquaintance chains created
by the students' choices. Rapport was interested in connectivity, i.e. the
fraction of the total population that would be contacted by tracing friendship choices from an arbitrary starting population of nine individuals.
Rapoport and his associates (Rapoport and Horvath, 1961; Foster et al.,
1963; Rapoport, 1953; 1963) have developed a mathematical model to
describe this tracing procedure. The model takes as a point of departure
random nets constructed in the following manner: a small number of points
is chosen from a larger population and a fixed number of "axones" is extended from each of these points to a set of target points chosen a t random
from the population. The same fixed number of axones is then extended from
each of the target points to a set of second generation target points, and
the process is repeated indefinitely. A target point is said to be of the
tth remove if it is of the tth generation and no lower generation. Rapoport
then suggests a formula for calculating the fraction, Pt, of the population
points which are targets of the tth remove. He is also able to extend the
formula to nonrandom nets, such as those created in the Rapport and
Horvath empirical study, by introducing a number of "biases" into the
random net model. Rapoport shows that two parameters, obtainable from the
data, are sufficient to produce a close fit between the predictions of the model
and the empirical outcome of the trace pr0cedure.l
Rapoport's model was designed to describe a trace procedure quite different from the one employed in the present study; however, it has some
relation to the small world problem. If we set the number of axones traced
from a given individual equal to the total number of acquaintances of an
average person, the Rapport model predicts the total fraction of the
population potentially traceable a t each remove from the start, serving
precisely the aims of the model of Pool and Kochen. (It should, however,
be noted that Rapoport's model deals with asymmetric nets, and it would
be difficult to modify the model to deal with general symmetric nets, which
characterize the small world phenomenon.)
Despite the goodness of fit between Rapoprt's model and the data from
1 There is additional empirical evidence (Fararo and Sunshine, 1964) and theoretical
support (Abelson, 1967) for the assumption that two parameters are sufficient to describe
the Rapoport tracing procedure, i.e. that more complex biases have minimal effects on
connectivity in friendship nets.
SOCIOMETRY two large sociograms, there are unsolved problems in the model, as
Rapoport himself and others (Fararo and Sunshine, 1964) have pointed
out. The Pool-Kochen model involves assumptions difficult for an empirically oriented social scientist to accept, such as the assumption that society
may be partitioned into a set of groups alike in size and in internal and
external connectedness. I n the absence of empirical data, it is difficult to
know which simplifying assumptions are likely to be fruitful. On the other
hand, with regard to the empirical study of Rapoport and Horvath, the
fact that the total population employed was small, well-defined, and homogeneous leaves open many questions about the nature of acquaintance nets
. ~ empirical study of American society as a whole may
in the larger s o ~ i e t yAn
well uncover phenomena of interest both in their own right and as constraints on the nature of any correct mathematical model of the structure
of large-scale acquaintanceship nets.
PROCEDURE
This paper follows the procedure for tracing acquaintance chains devised
and first tested by Milgram (1967). The present paper introduces an experimental variation in this procedure, by varying "starting populations";
it also constitutes a first technical report on the small world method.
The procedure may be summarized as follows: an arbitrary "target person"
and a group of "starting persons" were selected, and an attempt was made
to generate an acquaintance chain from each starter to the target. Each
starter was provided with a document and asked to begin moving it by mail
toward the target The document described the study, named the target, and
asked the recipient to become a participant by sending the document on.
It was stipulated that the document could be sent only to a first-name
acquaintance of the sender. The sender was urged to choose the recipient
in such a way as to advance the progress of the document toward the target;
several items of information about the target were provided to guide each
new sender in his choice of recipient. Thus, each document made its way
along an acquaintance chain of indefinite length, a chain which would end
only when it reached the target or when someone along the way declined
to participate. Certain basic information, such as age, sex and occupation,
was collected for each participant.
I n addition to the Pool-Kochen and Rapoport work, there are numerous other studies
of social network phenomena tangentially related to the small-world problem. Two wellknown examples are Bailey's The Mathematical Theory of Epidemics and Coleman, Katz
and Menzel's Medical Innovation. Bailey's work deals with diffusion from a structured
source, rather than with convergence on a target from a set of scattered sources, as in
the present study. The Coleman, Katz and Menzel study deals with an important substantive correlate of acquaintance nets, namely information diffusion.
STUDY OF THE SMALL WORLD PROBLEM
429
We were interested in discovering some of the internal structural features
of chains and in making comparisons across chains as well. Among the
questions we hoped to answer were the following: How many of the starters
--if any-would
be able to establish contact with the target through a
chain of acquaintances? How many intermediaries would be required to
link the ends of the chains? What form would the distribution of chain
lengths take? What degree of homogeneity in age, sex, occupation, and other
characteristics of participants would be observed within chains? How would
complete chains differ from incomplete on these and other dimensions?
An additional comparison was set up by using three distinct starting
subpopulations. The target person was a Boston stockbroker; two of the
starting populations were geographically removed from him, selected from
the state of Nebraska. A third population was selected from the Boston
area. One of the Nebraska groups consisted of bluechip stockholders, while
the second Nebraska group and the Boston group were "randomly" selected
and had no special access to the investment business. By comparisons across
these groups we hoped to assess the relative effects of geographical distance
and of contact with the target's occupational group. Moreover we hoped
to establish a strategy for future experimental extensions of the procedure,
in which the sociological characteristics of the starting and target populations
would be systematically varied in order to expose features of social structure.
The primary research questions, then, involved a test of the feasibility
and fruitfulness of the method as well as an attempt to discover some elementary features of real social nets. Several experimental extensions of the
procedure are already underway. A more detailed description of the current
method is given in the following sections.
PARTICIPANTS.
Starting Population. The starting population for the study
was comprised of 296 volunteers. Of these, 196 were residents of the state
of Nebraska, solicited by mail. Within this group, 100 were systematically
chosen owners of blue-chip stocks; these will be designated ('Nebraska
stockholders" throughout this paper. The rest were chosen from the population at large; these will be termed the "Nebraska random" group. In addition to the two Nebraska groups, 100 volunteers were solicited through an
advertisement in a Boston newspaper (the "Boston random" group). Each
member of the starting population became the first link in a chain of
acquaintances directed at the target person.
Intermediaries. The remaining participants in the study, who numbered
453 in all, were in effect solicited by other participants; they were acquaintances selected by previous participants as people likely to extend the chain
toward the target. Participation was voluntary. Participants were not paid,
nor was money or other reward offered as incentive for completion of chains.
SOCIOMETRY
430
THE DOCUMENT.
The 296 initial volunteers were sent a document which
was the principal tool of the in~estigation.~
The document contained:
a. a description of the study, a request that the recipient become a participant, and a set of rules for participation;
b. the name of the target person and selected information concerning him;
c. a roster, to which each participant was asked to affix his name;
d. a stack of fifteen business reply cards asking information about each
participant.
Rules for Participation. The document contained the following specific
instructions to participants:
a. Add your name to the roster so that the next person who receives this
folder will know whom it came from.
b. Detach one postcard from the bottom of this folder. Fill it out and
return it to Harvard University. No stamp is needed. The postcard
is very important. I t allows us to keep track of the progress of the
folder as it moves toward the target person.
c. If you know the target person on a personal basis, mail this folder
directly to him (her). Do this only if you have previously met the
target person and know each other on a first name basis.
d. If you do not know the target person on a personal basis, do not
try to contact him directly. Instead, mail this folder to a personal
acquaintance who is more likely than you to know the target person.
You may send the booklet on to a friend, relative, or acquaintance, but
it must be someone you know personally.
Target Person. The target person was a stockholder who lives in Sharon,
Massachusetts, a suburb of Boston, and who works in Boston proper. In
addition to his name, address, occupation and place of employment, participants were told his college and year of graduation, his military service
dates, and his wife's maiden name and hometown. One question under investigation was the type of information which people would use in reaching
the target.
Roster. The primary function of the roster was to prevent "looping," i.e.,
to prevent people from sending the document to someone who had already
received it and sent it on. An additional function of the roster was to motivate people to continue the chains. I t was hoped that a list of prior participants, including a personal acquaintance who had sent the document to
3A
photographic reproduction of this experimental document appears in Milgram,
1969: 110-11.
STUDY OF THE SMALL WORLD PROBLEM
43 1
the recipient, would create willingness on the part of those who received the
document to send it on.
Tracer Cards. Each participant was asked to return to us a business reply
card giving certain information about himself and about the person to whom
he sent the document. The name, address, age sex and occupation of the
sender and sender's spouse were requested, as were the name, address, sex
and age of the recipient. In addition, the nature of the relationship between
sender and recipient-whether
they were friends, relatives, business assoasked. Finally, participants were asked why they had
ciates, etc.-was
selected the particular recipient of the folder.
The business reply cards enabled us to keep running track of the progress
of each chain. Moreover, they assured us of getting information even from
chains which were not completed, allowing us to make comparisons between
complete and incomplete chains.
RESULTS COMPLETIONS.
217 of the 296 starting persons actually sent the document
on to friends. Any one of the documents could reach the target person only
if the following conditions were met: 1) recipients were sufficiently motivated
to send the document on to the next link in the chain; 2) participants were
able to adopt some strategy for moving the documents closer to the target
(this condition further required that the given information allow them to
select the next recipient in a manner that increased the probability of contacting the target); 3) relatively short paths were in fact required to link
starters and target (otherwise few chains would remain active long enough
to reach completion). Given these contingencies, there was serious doubt
in the mind of the investigators whether any of the documents, particularly
those starting in an area remote from the target person, could move through
interlocking acquaintance networks and converge on him. The actual outcome was that 64 of the folders, or 29 per cent of those sent out by starting
persons, eventually reached the target.
DISTRIBUTION
OF CHAINLENGTHS.Complete Chains. Figure 1 shows the
frequency distribution of lengths of the completed chains. "Chain length"
is here defined as the number of intermediaries required to link starters and
target. The mean of the distribution is 5.2 links.
I t was unclear on first inspection whether the apparent drop in frequency
a t the median length of five links was a statistical accident, or whether the
distribution was actually bimodal. Further investigation revealed that the
summary relation graphed in Figure 1 concealed two underlying distributions: when the completed chains were divided into those which approached
the target through his hometown and those which approached him via
SOCIOMETRY NUMBER OF INTERMEDIARIES
FIGURE 1
Lengths of Completed Chains
Boston business contracts, two distinguishable distributions emerged. The
mean of the Sharon distribution is 6.1 links, and that of the Boston distribution is 4.6. The difference is significant a t a level better than .0005, as
assessed by the distribution-free Mann-Whitney U test. (Note that more
powerful statistical tests of the significance of differences between means
cannot be applied to these data, since those tests assume normality of underlying distributions. The shape of the true or theoretical distribution of
lengths of acquaintance chains is precisely what we do not know.)
Qualitatively, what seems to occur is this. Chains which converge on
the target principally by using geographic information reach his hometown
or the surrounding areas readily, but once there often circulate before entering the target's circle of acquaintances. There is no available information
to narrow the field of potential contacts which an individual might have
within the town. Such additional information as a list of local organizations
STUDY OF THE SMALL WORLD PROBLEM
433
of which the target is a member might have provided a natural funnel,
facilitating the progress of the document from town to target person. By
contrast, those chains which approach the target through occupational
channels can take advantage of just such a funnel, zeroing in on him first
through the brokerage business, then through his firm.
Incomplete Chains. Chains terminate either through completion or dropout: each dropout results in an incomplete chain. Figure 2 shows the number
of chains which dropped out at each "remove" from the starting population.
The "0th remove" represents the starting population itself: the "first remove" designates the set of people who received the document directly from
members of the starting population. The "second remove" received the
document from the starters via one intermediary, the third through two
intermediaries, etc. The length of an incomplete chain may be defined as
the number of removes from the start a t which dropout occurs, or, equivalently, as the number of transmissions of the folder which precede dropout. By this definition, Figure 2 represents a frequency distribution of the
lengths of incomplete chains. The mean of the distribution is 2.6 links.
The proportion of chains which drop out at each remove declines as
REMOVE FROM START AT WHICH TERMINATION OCCURRED FIGURE
2
Lengths o f Incomplete Chains
chains grow in length, if that proportion is based on all chains active at each
remove (those destined for completion as well as incompletion). About 2 7
per cent of the 296 folders sent to the starting population are not sent on.
Similarly, 27 per cent of the 217 chains actually initiated by the starters
die at the first remove. The percentage of dropouts then appears to fall.
I t also begins to fluctuate, as the total number of chains in circulation grows
small, and an increasing proportion of completions further complicates the
picture.
I t was argued earlier that, in theory, any two people can be linked by
a t least one acquaintance chain of finite length, barring the existence of
totally isolated cliques within the population under study. Yet, incomplete
chains are found in our empirical tracing procedure because a certain proportion of those who receive the document do not send it on. It is likely that
this occurs for one of two major reasons: 1) individuals are not motivated
to participate in the study; 2 ) they do not know to whom to send the
document in order to advance it toward the target.
For purposes of gauging the significance of our numerical results, it would
be useful to know whether the dropouts are random or systematic, i.e.,
whether or not they are related to a chain's prognosis for rapid completion.
I t seems possible, for example, that dropouts are precisely those people who
are least likely to be able to advance the document toward the target. If
SO, the distribution of actual lengths of completed chains would understate
the true social distance between starters and target by an unknown amount.
(Even if dropouts are random, the observed distribution understates the
true distribution, but by a potentially calculable amount.) We can offer
some evidence, however, that this effect is not powerful.
First, it should be clear that, though people may drop out because they
see little possibility that any of their acquaintances can advance the folder
toward the target, their subjective estimates are irrelevant to the question
just raised. Such subjective estimates may account for individual decisions
not to participate; they do not tell us whether chains that die in fact would
have been longer than others had they gone to completion. People have poor
intuitions concerning the lengths of acquaintance chains. Moreover, people
can rarely see beyond their own acquaintances; it is hard to guess the circles
in which friends of friends-not to mention people even more remotely connected to oneself-may move.
More direct evidence that dropouts may be treated as "random" can be
gleaned from the tracer cards. I t will be recalled that each participant was
asked for information not only about himself but also about the person
to whom he sent the document. Thus some data were available even for
dropouts, namely age, sex, the nature of their relationship to the people
TABLE
I
Activity of Chains at Each Remove
All Chains
Remove
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Chains Reaching
Completions
this Remove
a t this Remove
296
217
158
122
95
76
50
38
17
10
6
4
1
1
1
0
0
2
3
8
14
8
16
6
2
2
3
0
0
0
Incomplete Chains Only
Dropouts
at this Remove
79
59
34
24
11
12
4
5
1
2
0
0
0
0
1
Per cent
Dropouts
27
27
22
20
12
16
8
13
6
20
0
0
0
0
100
Remove
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Chains Reaching
this Remove
Dropouts a t
this Remove
232
153
94
60
36
25
13
9
4
3
1
1
1
1
1
79
59
34
24
11
12
4
5
1
2
0
0
0
0
1
Per cent
Dropouts
34
39
36
40
31
48
31
55
25
67
0
0
0
0
100
SOCIOMETRY preceding them in the chain, and the reason the dropout had been selected
to receive the document. These four variables were tabulated for dropouts
versus non-dropouts. None of the resulting contingency tables achieved the
.05 level of statistical significance by chi-square test; we are therefore led
to accept the null hypothesis of no difference between the two groups, a t
least on this limited set of variables. Of course, a definitive answer to the
question of whether dropouts are really random must wait until the determinants of chain length are understood, or until a way is found to force
all chains to ~ompletion.~
SUBPOPULATION
COMPARISONS.
A possible paradigm for future research
using the tracing procedure described here involves systematic variation of
the relationship between the starting and target populations. One such study,
using Negro and White starting and target groups, has already been completed by Korte and Milgram (in press). In the present study, which involved
only a single target person, three starting populations were used (Nebraska
random, Nebraska stockholders, and Boston random.) The relevant experimental questions were whether the proportion of completed chains or mean
chain lengths would vary as a function of starting population.
Chain Length. Letters from the Nebraska subpopulations had to cover a
geographic distance of about 1300 miles in order to reach the target, whereas
letters originating in the Boston group almost all started within 25 miles
of his home and/or place of work. Since social proximity depends in part
on geographic proximity, one might readily predict that complete chains
originating in the Boston area would be shorter than those originating in
Nebraska. This presumption was confirmed by the data. As Table 2 shows,
chains originating with the Boston random group showed a mean length
of 4.4 intermediaries between starters and target, as opposed to a mean
length of 5.7 intermediaries for the Nebraska random group. (p-L.001 by
4 Professor Harrison White of Harvard University has developed a technique for adjusting raw chain length data to take account of the dropout problem. His method assumes
that dropouts are "random," in the following sense. An intermediary who knows the
target sends him the folder, completing the chain, with probability 1. Otherwise, an
intermediary throws away the folder with fixed probability 1-a, or sends it on with
probability a. If sent on, there is a probability Qi (which depends on number of removes
from the origin) that the next intermediary knows the target. The data is consistent with
a value for a of approximately 0.75, independent of remove from the origin, and hence
with a "random" dropout rate of 25 per cent. The limited data further suggest that Qi
grows in a "staircase" pattern from zero (at zero removes from the starting population) to
approximately one-third a t six removes, remaining constant thereafter. Based on these
values, the hypothetical curve of completions with no dropouts resembles the observed
curve shifted upward; the median length of completed chains rises from 5 to 7, but no
substantial alteration is required in conclusions drawn from the raw data.
SOCIOMETRY
438
a one-tailed Mann-Whitney U test.) Chain length thus proved sensitive to
one demographic variable-place of residence of starters and target.
The Nebraska stockholder group was presumed to have easy access to
contacts in the brokerage business. Because the target person was a stockbroker, chains originating in this group were expected to reach the target
more efficiently than chains from the Nebraska random group. The chainlength means for the two groups, 5.7 intermediaries for the random sample
and 5.4 for the stockholders, differed in the expected direction, but the
difference was not statistically significant by the Mann-Whitney test. The
stockholders used the brokerage business as a communication channel more
often than did the random group; 60.7 per cent of all the participants in
chains originating with the stockholder group reported occupations connected
with finance, while 31.8 per cent of participants in chains originating in
the Nebraska random group were so classified.
Proportion of Completions. As indicated in Table 3, the proportions of
chains completed for the Nebraska random, Nebraska stockholder, and
Boston subpopulations were 24 per cent, 31 per cent and 35 per cent,
respectively. Although the differences are not statistically significant, there
is a weak tendency for higher completion rates to occur in groups where
mean length of completed chains is shorter. This result deserves brief
discussion.
Let us assume that the dropout rate is constant at each remove from the
start. If, for example, the dropout rate were 25 per cent then any chain
would have a 75 per cent probability of reaching one link, (.75)= of reaching
two links, etc. Thus, the longer a chain needed to be in order to reach completion, the less likely that the chain would survive long enough to run its full
course. In this case, however, chain-length differences among the three groups
were not sufficiently large to produce significant differences in completion
rate. Moreover, if the dropout rate declines as chains grow long, such a decrease would off-set the effect just discussed and weaken the observed inverse
relation between chain length and proportion of completions.
TABLE 3
Proportion of Completions for Three Starting Populations
Starting Population
Nebraska Random
Complete
Incomplete
18
58
76
(24%)
(76%)
(100%)
x1=2.17, df.=2, $>.3, N.S.
Nebraska Stock.
24
54
78
(31%)
(69%)
(100%)
Boston
22
41
63
(35%)
(65%)
(100%)
Total
64
153
217
(29%)
(71%)
(100%)
STUDY OF THE SMALL WORLD PROBLEM
439
C ~ B ~ MCHANNELS.
ON
AS chains converge on the target, common channels
appear-that is, some intermediaries appear in more than one chain. Figure
3 shows the pattern of convergence. The 64 letters which reached the target
were sent by a total of 26 people. Sixteen, fully 25 per cent, reached the
target through a single neighbor. Another 10 made contact through a single
business associate, and 5 through a second business associate. These three
"penultimate links" together accounted for 48 per cent of the total completions. Among the three, an interesting division of labor appears. Mr. G,
64 INDIVIDUALS 55
INDIVIDUALS
5 CHAINS
2 CHAINS
--
-
-
26
INDIVIDUALS
16 CHAINS
L__1
FIGURE 3 Common Paths Appear as Chains Converge on the Target SOCIOMETRY
who accounted for 16 completions, is a clothing merchant in the target's
hometown of Sharon; Mr. G funnelled toward the target those chains which
were advancing on the basis of the target's place of residence. Twenty-four
of
chains reached the target from his hometown; Mr. G accounted for
those completions. All the letters which reached Mr. G came from residents
of Sharon. By contrast, Mr. D and Mr. P, who accounted for 10 and 5
completions, respectively, were contacted by people scattered around the
Boston area, and in several cases, by people living in other cities entirely.
On the other hand, whereas Mr. G received the folder from Sharon residents
in a wide variety of occupations, D and P received it almost always from
stockbrokers. A scattering of names appear two or three times on the
list of penultimate links; seventeen names appear once each.
Convergence appeared even before the penultimate link. Going one step
further back, to people two removes from the target, we find that the 64
chains passed through 55 individuals. One man, Mr. B, appeared 5 times,
and on all occasions sent the document to Mr. G. Other individuals appeared
two or three times each.
Eighty-six per cent of the parOF
ADDITIONAL
CHARACTERISTICSCHAINS.
ticipants sent the folder to persons they described as friends and acquaintances; 14 per cent sent it to relatives. The same percentages had been
observed in an earlier pilot study.
Data on patterns of age, sex and occupation support the plausible hypothesis that participants select recipients from a pool of individuals similar
to themselves. The data on age support the hypothesis unequivocally; the
data on sex and occupation are complicated by the characteristics of the
target and the special requirement of establishing contact with him.
Age was bracketed into ten-year categories and the ages of those who
sent the document tabled against the ages of those to whom they sent it.
On inspection the table showed a strong tendency to cluster around the
diagonal, and a chi-square test showed the association to be significant at
better than the ,001 level.
Similarly, the sex of each sender was tabled against the sex of the
corresponding recipient. Men were ten times more likely to send the document to other men than to women, while women were equally likely to
send the folder to males as to females (p<.001).These results were affected
by the fact that the target was male. I n an earlier pilot study using a
female target, both men and women were three times as likely to send the
document to members of the same sex as to members of the opposite sex.
Thus there appear to be three tendencies governing the sex of the recipient:
(1)there is a tendency to send the document to someone of one's own sex,
but ( 2 ) women are more likely to cross sex lines than men, and (3) there
STUDY OF THE SMALL WORLD PROBLEM
44 1
is a tendency to send the document to someone of the same sex as the
target person.
The occupations reported by participants were rated on two components--one of social status and one of "industry" affiliation, that is,
the subsector of the economy with which the individual would be likely
to deal. The coding system was ad hoc, designed to fit the occupational
titles supplied by participants. Tabling the status and "industry" ratings
for all senders of the document against those of respective recipients, we
observed a strong tendency for people to select recipients similar to
themselves on both measures (p<.001 for both tables). However, the
strength of the relationship for industry seemed to be largely due to a
tendency for the folder to stay within the finance field once it arrived
there, obviously because the target was affiliated with that field. Moreover, the participants in the study were a heavily middle-class sample,
and the target was himself a member of that class. Thus there was no
need for the document to leave middle-class circles in progressing from
starters to target.
When separate contingency tables were constructed for complete and incomplete chains, the above results were obtained for both tables. Similarly,
when separate tables were constructed for chains originating in the 3 starting
populations, the findings held up in all 3 tables. Thus, controlling for completion of chains or for starting population did not affect the finding of
demographic homogeneity within chains.
CONCLUSIONS
The contribution of the study lies in the use of acquaintance chains to
extend an individual's contacts to a geographically and socially remote target,
and in the sheer size of the population from which members of the chains
were drawn. The study demonstrated the feasibility of the "small world"
technique, and took a step toward demonstrating, defining and measuring
inter-connectedness in a large society.
The theoretical machinery needed to deal with social networks is still in
its infancy. The empirical technique of this research has two major contributions to make to the development of that theory. First, it sets an
upper bound on the minimum number of intermediaries required to link
widely separated Americans. Since subjects cannot always foresee the most
efficient path to a target, our trace procedure must inevitably produce chains
longer than those generated by an accurate theoretical model which takes
full account of all paths emanating from an individual. The mean number
of intermediaries observed in this study was somewhat greater than five;
442
SOCIOMETRY
additional research (by Korte and Milgram) indicates that this value is
quite stable, even when racial crossover is introduced. Both the magnitude
and stability of the parameter need to be accounted for. Second, the study
has uncovered several phenomena which future models should explain. In
particular, the convergence of communication chains through common individuals is an important feature of small world nets, and it should be
accounted for theoretically.
There are many additional lines of empirical research that may be examined with the small world method. As suggested earlier, one general
paradigm for research is to vary the characteristics of the starting person
and the target. Further, one might systematically vary the information provided about the target in order to determine, on the psychological side, what
strategies people employ in reaching a distant target, and on the sociological
side, what specific variables are critical for establishing contact between
people of given characteristics.
REFERENCES
Abelson, R. P.
1967 "Mathematical models in social psychology." Pp. 1-54 in L. Berkowitz (ed.)
Advances in Experimental Social Psychology, Vol. 111. New York: Academic
Press.
Bailey, N. T. J.
The Mathematical Theory of Epidemics. New York: Hafner.
1957
Coleman, J. S., E. Katz and H. Menzel
Medical Innovation: A Diffusion Study. Indianapolis: Bobbs-Merrill.
1966
Fararo, T. J. and M. H. Sunshine
A Study of a Biased Friendship Net. Syracuse: Youth Development Center,
1964
Syracuse University.
Foster, C. C., A. Rapoport and C. J. Orwant
"A study of a large sociogram 11. Elimination of free parameters." Behavioral
1963
Science 8( January) :56-65.
Gurevitch, M.
1961
The Social Structure of Acquaintanceship Networks. Unpublished doctoral
dissertation, Cambridge: M.I.T.
Korte, C. and S. Milgram
Acquaintance Links Between White and Negro Populations: Application of
the Small World Method. Journal of Personality and Social Psychology
(in press).
Milgram, S.
1967
"The small world problem." Psychology Today l(May):61-67.
1969
"Interdisciplinary thinking and the small world problem." Pp. 103-120 in
Muzafer Sherif and Carolyn W. Sherif (eds.) Interdisciplinary Relationships
ip the Social Sciences. Chicago: Aldine Publishing Company.
STUDY OF THE SMALL WORLD PROBLEM
443
Pool, I. and M. Kochen
A Non-Mathematical Introduction to a Mathematical Model. Undated
mimeo. Cambridge: M.I.T.
R a p p o r t , A.
"Spread of information through a population with socio-structural bias."
1953
Bulletin of Mathematical Biophysics 15(December):523-543.
"Mathematical models of social interaction." Pp. 493-579 in R. D. Luce,
1963
R. R. Bush and E. Galanter (eds.) Handbook of Mathematical Psychology,
Vol. 11. New York: John Wiley and Sons.
Rapoport, A. and W. J. Horvath "A study of a large sociogram." Behavioral Science 6(0ctober):279-291. 1961
Download